Running Claude Code on Amazon Bedrock

Anthropic's Claude Code agentic coding tool can now route through Amazon Bedrock instead of Anthropic's API. This matters because enterprise teams previously blocked from using AI coding tools due to data residency, compliance, or billing constraints now have a viable on-ramp. The integration ensures code and prompts never leave a team's AWS account, telemetry to Anthropic is disabled by default, and all usage flows through existing IAM, CloudTrail, and AWS billing infrastructure. The video details several operational mechanics: cross-region inference to handle throttling by routing requests across regions within a geography; the newer Mantle endpoint for native API compatibility, separate input/output token quotas, and zero operator access; and prompt caching that can reduce costs up to 90% with an optional 1-hour cache extension. For engineering managers, the practical sections cover granular cost attribution via IAM principals and application inference profiles, letting teams track spend per developer, project, or model. Bedrock Guardrails enable centralized content filtering and PII redaction that can be committed to a shared project config, enforcing consistent governance. The tradeoff is that some advanced features available in Anthropic's direct API are not yet supported on Bedrock, though the Mantle endpoint closes several gaps. The underlying shift isn't about a new tool feature—it's that agentic coding is becoming procurement-able infrastructure, which changes who controls adoption and how it scales across an organization. Most teams will focus on the setup steps. The real signal is that compliance, cost attribution, and governance are now configurable artifacts in a JSON file, not organizational blockers.